AI Safety & Disclosure Policy Last Updated: June 14, 2026

1. Overview Phasely includes AI-powered features collectively referred to as the “AI Coach,” which provide educational information related to menstrual cycle awareness, fasting, nutrition, exercise, and general wellness. This document explains how the AI Coach works, its limitations, and the safety measures in place to protect users. This policy supplements our Terms of Service, Privacy Policy, and Medical Disclaimer. In case of conflict between this policy and the Terms of Service, the Terms of Service prevail.
2. Nature of the AI Coach The AI Coach is an artificial intelligence system designed to: Provide educational explanations based on curated wellness and health knowledge. Summarize information from scientific and expert sources. Offer general guidance on fasting, nutrition, and cycle-based wellness practices. Support users in understanding their menstrual cycle phases. The AI Coach does NOT: Provide medical advice, diagnosis, or treatment. Replace consultation with healthcare professionals. Guarantee accuracy, completeness, or suitability of responses. Make personalized medical decisions.
3. Educational Purpose Only All responses generated by the AI Coach are for educational and informational purposes only. They should not be interpreted as medical advice, clinical recommendations, prescriptions, or treatment plans. Users are responsible for verifying any health-related information with a qualified healthcare professional.
4. Known Limitations of AI Users acknowledge that AI systems may: Produce incorrect or incomplete information. Generate responses that appear confident even when uncertain (“hallucination”). Be influenced by patterns in training data that may include biases. Misinterpret user inputs. Generate outdated or generalized responses. Lack awareness of individual medical history or conditions. Vary in response quality based on how questions are phrased. Have limited ability to assess context across long conversations. Be unable to verify the accuracy of claims about your individual health status. Fail to recognize when a question requires human judgment. Reflect uncertainty or disagreement in scientific literature. Phasely does not guarantee the accuracy, reliability, or completeness of AI-generated content.
5. Safety Guardrails Phasely implements safety systems designed to reduce the risk of harm. 5.1 Retrieval-Based Grounding The AI Coach uses Retrieval-Augmented Generation (RAG), meaning responses are grounded in a curated knowledge base of expert sources rather than generated purely from the underlying language model. This is intended to reduce the likelihood of fabricated or unsupported claims. All retrieved sources are visible to users through citation chips, allowing users to evaluate the basis for any recommendation. 5.2 Multi-Stage Safety Routing Each question submitted to the AI Coach passes through automated safety checks before generating a response: Safety classifier: Detects questions involving warning signs, medical emergencies, or scenarios requiring professional referral. Disagreement detection: Identifies questions where expert sources disagree and surfaces multiple perspectives. Standard retrieval: Returns relevant educational content with citations. When safety conditions are detected, the AI Coach is instructed to recommend consultation with a healthcare professional and may decline to provide specific guidance. 5.3 Medical and High-Risk Content The AI Coach is designed to: Decline requests for diagnosis of medical conditions. Avoid prescribing medications, dosages, or specific supplement regimens. Avoid recommending extreme fasting protocols. Avoid providing guidance on weight loss as a primary goal. Encourage consultation with healthcare professionals when questions suggest medical concerns. Use language that is supportive rather than diagnostic. 5.4 High-Risk User Groups Users who indicate any of the following conditions during onboarding receive conservative guidance from the AI Coach, or, in some cases, are advised to consult their healthcare provider before continuing with the application: Pregnancy, breastfeeding, or planning pregnancy. History or current symptoms of eating disorders. Diabetes requiring medication or insulin. Untreated thyroid disorders or active autoimmune conditions. Cardiovascular conditions or blood pressure disorders. Medications requiring consistent food intake. Hormonal contraception that suppresses natural menstrual cycles. Any condition for which a healthcare provider has advised against fasting. 5.5 System Prompt and Voice Constraints The AI Coach operates with a defined set of system instructions that constrain its voice and behavior. These instructions explicitly prohibit: Providing absolute or universal statements about safety (“this is safe”). Giving specific medical or dosing recommendations. Adopting a diagnostic tone. Generating bottom-line summaries that oversimplify nuanced topics. Claiming authority that exceeds the underlying educational sources.
6. Scientific Disagreement and Source Transparency Phasely acknowledges that scientific and expert opinions on fasting, cycle syncing, hormonal health, and metabolic health are not fully unified. Knowledge in these areas is evolving. When responding to questions where expert sources disagree, the AI Coach may: Present multiple perspectives clearly. Identify which sources hold which views. Distinguish between peer-reviewed research, clinical practice, and practitioner-led frameworks. Note where evidence is limited or contested. Encourage critical evaluation of recommendations. Phasely’s knowledge base draws from multiple expert “camps” with different perspectives, including authors and practitioners specializing in cycle-aware fasting, sports science for women, clinical hormone health, perimenopause and menopause, and evidence-based skeptical analysis. Citations from the AI Coach indicate which sources are being referenced.
7. Knowledge Base Curation The AI Coach draws from a curated knowledge base of expert sources. This knowledge base is: Manually curated — content is selected, structured, and reviewed by Phasely. Subject to ongoing verification — during beta and early production phases, knowledge base content is cross-referenced against original published sources. Updated over time — content may be added, modified, or removed as scientific understanding evolves or sources are verified. Transparent in attribution — citations identify the source of each piece of information. Knowledge base content reflects published expert opinions and research as of the curation date. We do not claim that all content in the knowledge base has been independently scientifically validated, and some content may represent the views of specific authors or practitioners rather than scientific consensus. If you believe a citation is misattributed, taken out of context, or factually incorrect, please report it via hello@phasely.today.
8. No Emergency Use The AI Coach is not intended for use in medical emergencies. If you experience a medical emergency or serious symptoms, contact emergency services immediately and do not rely on Phasely for guidance. Emergency numbers: Switzerland: 144 (medical), 112 (general emergency) European Union: 112 United States: 911
9. Data Usage in AI Processing To provide responses, your inputs are processed by third-party AI providers. Providers used: Anthropic (anthropic.com) — for AI Coach response generation (Claude models) and meal photo analysis. Based in the United States. OpenAI (openai.com) — for text embeddings used in knowledge base retrieval. Based in the United States. What is sent to Anthropic: The text messages you type to Coach. Any meal photos you send to Coach. Your current cycle phase and day number. Lifestyle context you have shared (activity level, stress level, sleep quality). Any health conditions you disclosed during onboarding (e.g. cycling/perimenopausal/postmenopausal status). Recent conversation history for context (last several messages). What is sent to OpenAI: A small text excerpt of your query to generate a semantic embedding for knowledge base search. What is NOT sent to either provider: Your name, email, or contact information. Your complete cycle history. Your full account profile. Identifying information about you as a user. How providers use this data: Anthropic processes your inputs only to generate Coach’s response. Anthropic does NOT use your messages to train their AI models. OpenAI does NOT use API embedding inputs to train their models (per OpenAI’s enterprise terms). Phasely does NOT sell, rent, or share your data with advertisers, data brokers, or analytics providers. Your consent: Before Coach sends any data to these providers, you must explicitly consent within the Phasely app. You can withdraw consent at any time by contacting hello@phasely.today and requesting account deletion. Data retention by AI providers: Anthropic and OpenAI retain API request data per their own retention policies (currently 30 days for safety monitoring under their API Terms). Only the minimum data necessary to generate responses is shared with third-party providers. Sensitive personal identifiers should not be shared with the AI Coach. Third-party provider privacy policies: Anthropic: https://www.anthropic.com/legal/privacy OpenAI: https://openai.com/privacy
10. Reporting Concerns If you encounter an AI Coach response that you believe is harmful, factually incorrect, misattributed, or misleading about medical matters, please report it through: The in-app feedback mechanism (thumbs-down on individual responses). Email: hello@phasely.today with the subject line “AI Safety Report”. We review all safety reports to improve safety systems, update the knowledge base, or adjust system instructions. We aim to acknowledge safety reports within 7 days and address material concerns within 30 days.
11. User Responsibility By using the AI Coach, you acknowledge that you are solely responsible for decisions based on AI outputs, that you will use judgment when applying any recommendations, and that you understand that AI is an informational tool, not a medical authority.
12. Safety-First Design Principle Phasely is designed with a conservative health safety approach: When evidence is uncertain, the AI defaults to lower-risk recommendations. Extreme protocols (very long fasts, aggressive caloric restriction, intense daily training) are avoided in AI responses. Health and safety considerations are prioritized over optimization outcomes or aspirational goals. Citations are surfaced to enable users to evaluate sources, rather than presenting AI responses as final authority.
13. Regulatory Awareness Phasely’s AI Coach is classified as an AI system providing educational and informational content. It is not classified as a medical device or a high-risk AI system under current European Union AI Act categorization. We monitor developments in AI regulation, including the European Union AI Act, Swiss data protection laws, and AI governance frameworks.
14. Updates to This Policy We may update this AI Safety & Disclosure Policy as the product evolves, AI capabilities change, or regulatory requirements develop. When material changes occur, we will update the “Last Updated” date at the top of this policy and provide reasonable notice within the app.
15. Contact For questions about AI safety, to report safety concerns, or about this policy: Phasely Karolina Bittman Zürich, Switzerland Email: hello@phasely.today Please use these subject lines for appropriate routing: “AI Safety Report” — for reporting potentially harmful or inaccurate AI responses “AI Safety Question” — for general questions about this policy “Knowledge Base Feedback” — for reporting citation issues or knowledge concerns